ReGO: Reference-Guided Outpainting for Scenery Image

pytorch implementation of ReGO: Reference-Guided Outpainting for Scenery Image. Note: we only release the code of our Best model: BDIE with ReGO equipped i.e., ReGO_BDIE.

The framework of our ReGO:

Our proposed Adaptive Content Selective Moduel(ACS):

Examplar outputs of our model:

Requirements

pytorch
torchvision
torchsummary
numpy
Pillow
random
glob

Prepare datasets

1. Download NS6K and NS8K dataset
2. unzip the zip file to 'data' directory: unzip -d data NS6K+NS8K.zip

Train, run

CUDA_VISIBLE_DEVICES=GPU_ID python train.py --dataset_name ./data/NS6K_Train/ --batch_size BATCH_SIZE --n_epochs TRAIN_EPOCHES --gpu 0

or edit the train.sh file and run 'sh train.sh'

Test

#test and generate the images
CUDA_VISIBLE_DEVICES=GPU_ID python test.py --model MODEL_PATH --image-path ./data/NS6K_Test/ --output OUTPUT_DIR  --gpu 0 --use_gpu

#calculate FID
python fid.py GT_PATH OURPUR_DIR --gen_mode 'gen' --gpu GPU_ID

#calculate IS
CUDA_VISIBLE_DEVICES=GPU_ID python eval_fid_is_score.py --gen_image_path OUTPUT_DIR --use_gpus 0 --gen_mode 'gen' 

or edit the eval.sh and run sh eval.sh

Acknowledge

This project is built on the top of this code: Boundless. Thanks the authors' contribution.